Garcia-Martinez Alvaro, Vicente-Samper Jose María, Sabater-Navarro José María
Systems and Automatics Engineering Department, Miguel Hernández University, Avinguda de la Universitat d'Elx, Elche, 03202, Spain.
Systems and Automatics Engineering Department, Miguel Hernández University, Avinguda de la Universitat d'Elx, Elche, 03202, Spain.
Artif Intell Med. 2017 May;78:55-60. doi: 10.1016/j.artmed.2017.06.002. Epub 2017 Jun 10.
On occasions, a surgical intervention can be associated with serious, potentially life-threatening complications. One of these complications is a haemorrhage during the operation, an unsolved issue that could delay the intervention or even cause the patient's death. On laparoscopic surgery this complication is even more dangerous, due to the limited vision and mobility imposed by the minimally invasive techniques.
In this paper it is described a computer vision algorithm designed to analyse the images captured by a laparoscopic camera, classifying the pixels of each frame in blood pixels and background pixels and finally detecting a massive haemorrhage. The pixel classification is carried out by comparing the parameter B/R and G/R of the RGB space colour of each pixel with a threshold obtained using the global average of the whole frame of these parameters. The detection of and starting haemorrhage is achieved by analysing the variation of the previous parameters and the amount of pixel blood classified.
When classifying in vitro images, the proposed algorithm obtains accuracy over 96%, but during the analysis of an in vivo images obtained from real operations, the results worsen slightly due to poor illumination, visual interferences or sudden moves of the camera, obtaining accuracy over 88%. The detection of haemorrhages directly depends of the correct classification of blood pixels, so the analysis achieves an accuracy of 78%.
The proposed algorithm turns out to be a good starting point for an automatic detection of blood and bleeding in the surgical environment which can be applied to enhance the surgeon vision, for example showing the last frame previous to a massive haemorrhage where the incision could be seen using augmented reality capabilities.
有时，外科手术可能会引发严重的、潜在危及生命的并发症。其中一种并发症是手术期间出血，这是一个尚未解决的问题，可能会延迟手术，甚至导致患者死亡。在腹腔镜手术中，由于微创技术带来的视野和操作空间有限，这种并发症更加危险。
本文描述了一种计算机视觉算法，旨在分析腹腔镜摄像头拍摄的图像，将每一帧的像素分类为血液像素和背景像素，最终检测出大量出血情况。像素分类是通过将每个像素的RGB空间颜色参数B/R和G/R与使用这些参数的整帧全局平均值获得的阈值进行比较来实现的。出血的检测和起始点是通过分析先前参数的变化和分类的血液像素数量来实现的。
在对体外图像进行分类时，所提出的算法获得了超过96%的准确率，但在分析从实际手术中获取的体内图像时，由于光照不佳、视觉干扰或摄像头的突然移动，结果略有恶化，准确率超过88%。出血的检测直接取决于血液像素的正确分类，因此分析的准确率为78%。
所提出的算法是手术环境中血液和出血自动检测的一个良好起点，可用于增强外科医生的视野，例如显示大量出血前的最后一帧，在该帧中可以使用增强现实功能看到切口。
